Google Antigravity 2.0: A Standalone Desktop App Built Around Agents, Not Editors

Google announced Antigravity 2.0 at I/O 2026, and the headline detail is not a new feature or a model upgrade. It is a structural shift in what the product actually is. Antigravity is no longer a code editor with AI bolted on. It is a standalone desktop app built from the ground up around agent conversations, and the code editor is no longer the centrepiece.

That is a meaningful change in philosophy, and it has some practical consequences worth understanding.

What Antigravity 2.0 Actually Is

The new Antigravity desktop app ships alongside a command-line tool (the agy CLI), a Python/TypeScript/Go SDK, and a Managed Agents API. Google describes this as five surfaces sharing one agent harness. The idea is that whether you are working in the GUI, the terminal, or calling an API from your own application, you are running the same underlying agent with the same context and capabilities.

The default model powering all of this is Gemini 3.5 Flash, which Google says is 12x faster on Antigravity than on other surfaces due to platform-level optimisations. You can also use Gemini 3.1 Pro, Claude Sonnet 4.6, Claude Opus 4.6, or the open GPT-OSS-120B model if you prefer.

Subagents: The Part Worth Paying Attention To

The most significant capability shipping in early preview is subagent teamwork. The idea is straightforward: instead of a single agent working through a complex engineering task sequentially, Antigravity can now spin up multiple specialised subagents and run them in parallel.

Google says this is collapsing multi-day engineering efforts into hours. That is a bold claim, but they have backed it with a concrete example: their researchers used this approach to autonomously build a functional operating system using Gemini 3.5 Flash. That is the kind of demonstration designed to establish a ceiling, not a typical workflow.

For most developers, the near-term payoff will be more modest but still genuinely useful. Tasks that currently require you to context-switch across multiple tools or manually coordinate sequential steps become something you can hand off to an orchestrated set of agents. Think of it as assigning a small team rather than a single assistant.

The subagent system also includes a browser subagent that can spin up a controlled Chromium instance, interact with your web app, and return screenshots. If you have ever written a feature and then manually clicked through a UI to check it works, that is the kind of loop this is designed to close automatically.

What This Means for Enterprise Google Cloud Customers

This is the update that will matter most to teams already operating in Google Cloud. Enterprise customers can now connect Antigravity directly to their existing Google Cloud projects by logging in with Cloud OAuth and setting an Agent Platform Project ID and region.

Once connected, all agent inference runs via Agent Platform models inside your secure cloud boundary. That means Antigravity inherits your existing Google Cloud data privacy terms rather than requiring you to accept separate consumer terms or route data through a different service. Regional model endpoints give you control over where inference happens.

If your organisation has been interested in agentic development tooling but has been waiting for something that fits inside your existing compliance posture, this is the announcement to take back to your team.

A Note on the Gemini CLI

If you are currently using the standalone Gemini CLI, Google has confirmed it stops serving free and Pro requests on 18 June 2026. The replacement is the new agy CLI, which shares authentication, context, skills, and configuration with the Antigravity desktop app. Migration is worth prioritising before that deadline rather than after.

The Managed Agents API

Separate from the desktop app, Google is also launching Managed Agents in the Gemini API. A single API call provisions a remote Linux sandbox where an agent can reason, write and execute code, manage files, and browse the web. This is powered by the same Antigravity harness and is available in Google AI Studio for experimentation.

For developers building their own agent-powered products, this removes a significant chunk of infrastructure work. You do not need to manage sandboxing, tool execution environments, or orchestration logic yourself. You define your agent’s instructions and skills in markdown files and the harness handles the rest.

Pricing

Google is introducing a new AI Ultra plan at $100 per month, which includes a 5x higher usage limit in Antigravity compared to the existing AI Pro plan. If you are a heavy user of Antigravity for professional work, that headroom will likely matter more than the monthly cost.

The Practical Summary

Antigravity 2.0 is worth a serious look if you are a developer who has been using AI-assisted coding tools and wondering when the underlying model of “chat with a helpful autocomplete” would be replaced by something that can actually own a workflow. The subagent preview is early, but the direction is clear.

For enterprise teams on Google Cloud, the native project integration removes the main practical blocker that has kept tools like this in the “interesting experiment” category rather than standard tooling.

The full release is available now. The subagent teamwork capability is in early preview, meaning expect rough edges, but also an opportunity to understand how this changes your development patterns before it becomes the default.